|Abstract (english)|| |
Syntactic processing is a rapid and highly automatized process involved in the integration of new information during language comprehension and production (Van Gompel & Pickering, 2007). These processes proceed in an incremental fashion, i.e. once each word is encountered, it is integrated into the sentence analysis. Despite numerous research, there are still many remaining questions about the exact underlying mechanisms of syntactic processing, such as what drives the processor; which information a speaker of a language predominately relies on during processing and when; which linguistic and nonlinguistic factors are involved and whether there are crosslinguistic, intralinguistic and individual differences in the process (Papadopoulou, 2006). Inconsistencies in research methodology pose additional challenges in making firm conclusions, especially within a particular language. Therefore, there remains a need for conducting new experiments with more rigorous control of nonlinguistic (i.e., working memory capacity) and linguistic (i.e., structural-semantic) factors that would bring about novel findings based on the data in a range of languages, respecting their structural and other features. In search of the structures that could be systematically used in crosslinguistic studies of syntactic processing, ambiguous clauses started to stand out. Even though ambiguity can occur in all utterances that allow for multiple interpretations considering the linguistic and situational context (Van Gompel & Pickering, 2007: 290; Raffaelli, 2015: 267), ambiguous relative clauses (RC) are considered a standard as “RC attachment ambiguity occurs in a variety of languages and thus constitutes an ideal test case for investigating whether the human sentence processor varies from language to language” (Cuetos, Mitchell & Corley, 1996: 148). By analysing these sentences one can get the notion of linguistic factors that drive syntactic processing in the absence of a broader context (see also Raffaelli, 2015: 266-267). These are firm justifications for analysing the ways speakers of different languages interpret ambiguous RCs and whether they prefer attaching the clause to the first (N1) or to the second (N2) noun within the noun phrase (NP) when both are possible (Hemforth, Konieczny & Scheepers, 2000a). Attachment preference studies are mostly performed using offline and online methods, for example forcedchoice questionnaires and eye-tracking. This doctoral thesis is a psycholinguistic sentence processing study that investigates attachment preferences in Croatian, and is conducted within one of many sentence processing models, the construal (Frazier & Clifton, 1996). Construal hypothesis highlights the importance of structural-semantic relations of the two nouns within the NP preceding a subsequent relative clause. If the two nouns are in nonprepositional genitive forms, then none of them is thematically licensed and attachment can be influenced by other factors. Nevertheless, N1 attachment (or high attachment) is somewhat preferred as these two nouns belong to the same processing domain. On the other hand, if an NP preceding the relative clause constitutes a prepositional phrase, then the preposition is considered thematically-licensing, i.e. it can cause a focus shift and change of preference. Therefore, this model predicts N2 or low attachment across languages (Frazier & Clifton, 1996: 73). Since according to construal differences in syntactic processing and attachment preferences are explained by these types of structures (RCs with NP in non-prepositional genitive forms and thematically-licensing prepositional phrases) these are the structures finally involved in the current attachment preference doctoral research. Prior to the experimental part of the thesis, a critical overview of the syntactic processing models is provided, as well as the review of findings of similar experiments. After the results section, general discussion is given in a broader psycholinguistic context and in a language-specific one, with reference to the Croatian language and its structures. The main aim of the study was to investigate which structural-semantic factors of NP preceding RCs explain syntactic processing and speakers’ attachment preferences in the Croatian language. Additional aim was to deliver relevant findings for Croatian, and thus to contribute both to the general psycholinguistic perspective of the studied behaviours, and to the language-specific one. Within the construal model, three hypotheses were formed: H.1. Considering the first postulate of the construal and recent findings of relative clauses in Croatian, if the NP is in non-prepositional genitive form, N1 attachment preference is expected. H.2. Considering the second postulate of the construal, if the NP is a prepositional phrase in which the preposition is potentially thematically licensing and causes the focus shift, N2 attachment preference is expected. H.3. Considering the features of the Croatian syntax and former studies in Croatian, it is expected that, in the absence of morphological cues, animacy will facilitate processing and influence the change in attachment preference. The study has been conducted using both offline (a judgment task in the form of a forced-choice questionnaire) and online (eye-tracking) methods. Qualitative data have also been gathered and analysed based on some of the participants’ comments and answers. The questionnaire stimuli consisted of ambiguous relative clauses with both attachment sites equally plausible, and the eye-tracking stimuli consisted of both ambiguous and unambiguous sentences, i.e., those with experimental control of the attachment. The N1 and N2 attachment in those stimuli were forced by either grammatical, i.e. morphological means (gender agreement) or semantic means (semantic plausibility). There were also filler sentences which were completely unambiguous and randomly displayed to participants. Since several other factors can influence preferences besides the type of the NP, those have also been carefully controlled for. Stimuli were therefore constructed for the purpose of this doctoral study, and were not mere translations of stimuli used in other languages and studies. These additional control factors were as follows: 1) subject position of the NP; 2) length of the NP and RC; 3) frequency of the nouns within the NP; 4) number (singular) of the nouns within the NP; and 5) either the control or the manipulation with gender agreement and semantic plausibility of the structure. The validity of the stimuli has been examined and confirmed by independent participants who were not involved in the main study. Participants were monolingual speakers of the Croatian language, without the history of neurological or language deficits. Ninety-seven (97) were included in the first offline experiment, and forty-eight (48) of them were selected to participate in two subsequent online experiments. The selection was based on their working memory capacity (WMC), i.e. speakers with extremely low or extremely high WMC in relation to the rest of the group were excluded from the study. This was decided since some research point to the correlation of WMC and attachment preferences, but the exact nature of this correlation remains unresolved, and should therefore at least be controlled (Swets et al., 2007; Traxler, 2007; Payne et al., 2014, ...). The results were analysed quantitatively applying parametric statistics methods, and some of the participants’ comments have been analysed qualitatively. Questionnaire data consisted of coded answers (1 for N1 response and 0 for N2) and afterwards the percentages of N1 and N2 attachment preferences across participants have been calculated, for both type of NPs separately (genitive forms and prepositional phrases). If a participant decided to change the initial answer, his or her first choice was coded as it reflected the initial preference. Measures used in the analyses of online data were based on reading times and number of relevant events on the entire sentence region, as well as on initially selected areas of interest or AoIs (an NP segment and a disambiguation segment). The measures were thus as follows: total reading time (ms), total fixation duration (ms), first pass (ms), N of fixations, N of regressions. These eye-movements data have been gathered and preprocessed in the SMI BeGaze programme, and afterwards statistically analysed using rmANOVA and t-tests. Parametric statistics methods were used as the results were normally distributed. Repeated measures ANOVA was conducted for each of the mentioned measures on both AoIs, with two factors, each having two levels (2x2 design): 1) type of NP (non-prepositional genitive NP vs prepositional phrase NP) and attachment site (N1 and N2) for the first experiment, and 2) animacy (attachment to animate vs attachment to inanimate noun within the NP) and attachment site (N1 and N2). The results of the offline experiment point to the strong N1 attachment preference for both types of NPs. The percentage of N1 responses in the questionnaire was 70,3% for nonprepositional genitive forms and 81,1% for prepositional phrases within the NP. The online eye-tracking data in general showed longer reading times in stimuli with forced N2 than with forced N1 attachment for the entire sentence region and for the observed AoIs, indicating higher processing costs for N2 attachment in Croatian. Again, this happened irrespective of the type of NP. The second online experiment additionally revealed that animacy facilitates processing, nevertheless the attachment preference remained the same (N1). More specifically, statistically significant main effects have been obtained for both factors, but they were stronger for the attachment site than for the other factor in both experiments. Still, significant main effects of animacy have been obtained on multiple dependent variables (i.e., longer reading times and higher processing costs for the inanimate than the animate attachment conditions) in the second online experiment. Rather important is the finding that significant effects have been attained not only for the so called late measures (i.e., total reading times), but also for the early eye-tracking measures (i.e., first pass reading times), indicating several things: 1) the processor grasps relevant information in a rapid, automatized fashion; 2) attachment happens incrementally, but also 3) speakers rely on multiple levels of information early in the process. Qualitative data further corroborate these claims since they imply that speakers rely on multiple information when interpreting an utterance and making the sense of it. Together these results allowed for the acceptance of the H1, the rejection of the H2, and partial acceptance of the H3. To sum up the overall results, in the broader psycholinguistic perspective they add the following: - they refute some of the postulates of the construal thus emphasising its limitations; - they somewhat confirm the proposals of interactive models which favour the idea of rapid activation of multiple levels of analyses during reading / sentence processing / attachment and disambiguation. In the language-specific context they add the following: - they show that the information integration during RC processing in Croatian is somewhat more natural in the case of N1 attachment than in the case of N2 attachment; - they indicate that animacy can facilitate processing in Croatian – if the two antecedents (N1 and N2) differ in the animacy feature, processing will be eased if the subsequent clause modifies the animate antecedent. The results on the preferred N1 attachment are in line with previous attachment preference studies in Croatian, but also show that the reasons are not strictly related to prosodic cues, as previously reported (i.e., Lovrić, Bradley & Fodor, 2001; Lovrić, 2003). They are related also to the complex structural-semantic relations, as well as to the notion of dominance and prominence of the first mentioned referent in Croatian. The outcome that animacy facilitates processing, even though it does not lead to the change in attachment preference, is an important follow-up of the experiments on the factors that drive syntactic processing in Croatian, mostly agreement (Willer-Gold er al., 2016; Matić et al., 2019). Furthermore, the results support the rapid interaction of multiple levels (or cues) during syntactic processing – prosodic, structural, semantic, information (Erteschik-Shir, 2009), which are not easily distinguished or isolated. There are certain limitations to this study that should be taken into account while designing similar research in the future. Even though there is a firm theoretical and psycholinguistic rationale for using relative clauses when studying attachment preference across languages, despite their importance findings cannot be easily generalised to other structures. This is why studies of this sort should aim at investigating and analysing also other structures. Moreover, in this research working memory capacity task served as the means of homogenising the initial sample of participants, as it is known that different levels of working memory capacity can cause differences in attachment preferences. Still, in order to broaden psycholinguistic insights further research should aim at investigating the influence of working memory capacity on attachment preferences, i.e., to determine the exact direction of the influence. That might be important since the findings on this matter are not just different, but completely contradictory (Swets et al., 2007; Traxler, 2007). Finally, mainly young student population without language impairment has been involved in this study. It might be useful to examine how older speakers and speakers with language impairment interpret these types of structures, i.e. whether they rely on the same underlying mechanisms, when these processes happen and whether their attachment preferences differ from those of typical population.